Drug shortages: some possible project aims
In this post, I’d like to reflect a little on possible aims of the drug shortages project. These are the big questions I’ll be considering that can inform my analysis, whether I get to tackle them in this project or not. Please comment below with other ideas.
Why do drug shortages happen?
Obviously a complex question that, in part, deserves a revisit of our paper, as well as a dive into recent literature an a re-read of the Multi-Stakeholder Steering Committee on Drug Shortages report of 2017. That said, is there anything we can glean from the shortage reporting itself? Each report contains a stated “reason” for the shortage, although when I last looked at the data this appeared to be a low-information field: just a sentence selected from a list of 8 or so per-determined reasons. Nevertheless, perhaps worth looking at the trends in reported reasons over time.
Has anything changed?
I am painfully unaware of the current discussion and recent policy changes (if any) surrounding drug supply and shortages in Canada, so this is something I will need to read up on as I go. But there is still an interesting question here: since our paper was published, and mandatory reporting has gotten underway, has there been any change in the rate of drug shortages. (Ideally, you would think that reports of shortages should have nothing to do with the shortages themselves, but perhaps with increasing transparency and media attention that has come with mandatory reporting over the last 2+ years, drug companies are pressured to address issues causing the shortages?)
I can easily compare the trends in # of shortages over time as well as duration of shortages over time to see if there has been any aggregate change in shortages. Additionally, I can look at changes to the composition of shortages, e.g. changes to drug classes in shortage, and generics vs brand name.
Can we predict shortages?
I am not enthusiastic about the recent trend in the tech world to turn every problem into a prediction problem (thereby allowing the invoking the incantations of machine learning and AI) but it’s an intriguing idea that could be fruitful: to go fishing for regularities in the shortage reporting data by way of trying to predict upcoming shortages.
I suspect this would be an exercise in “feature engineering” – i.e. coming up with relevant features to put into a prediction model. So let’s brainstorm possible predictive features of drug shortage:
- Age of drug\
- Number of past shortages for a drug\
- Time since last shortage for a drug\
- Reason for past shortage\
- Shortages of other drugs made by the same manufacturer\
- Generic or brand name status\
- Existence of a competing generic version\
- etc..
Hmm.. Doing doing this work is an entire research project on its own, so I won’t get to it during my elective. Perhaps worth keeping in mind though.
How should we monitor shortages?
For those of us interested in the problem of drug shortages in Canada, what are meaningful metrics to monitor with respect to shortages that can help us understand the problem? Remember, I’m planning on making a dashboard website with regularly updated charts and tables showing the current state of the drug shortages?
Here are some possibilities:
- Current # of drugs in shortage
- Past trend of # of shortages (say, average over a week) or # of popular drugs in shortage
- Histogram of current shortages duration
- Estimated shortages as of next week
- List of drugs currently in shortage
- List of drugs anticipated to be in shortage next week
How do shortage reports change?
Annoyingly, shortage reports themselves can be updated over time as more information about the shortage becomes available. So, for example, the shortage report may be created initially with an estimated start date that later gets updated, and when the shortage itself is resolved the shortage status will change. It might be interesting to look at how shortage reports change over time in order to assess their utility (e.g. are shortage reports made early enough to be useful to consumers?) and accuracy (e.g. How good are manufacturers at estimating start/end dates? How often are these updated?)
Back when we first published our paper, the drug shortages database had no way to view the history of each shortage report. That is, it was not possible to see if and when a report was updated. It didn’t seem right that manufacturers were able to re-write history, so I spoke to the folks running the database to suggest they make the history available, and I also started a project to take regular snapshots of the entire database. You can find those snapshots here:
https://github.com/pipitone/drug-shortages-snapshots
But that project is no longer needed. It seems history for each shortage is now available on the database website! For example: